kstant0725

Lowering the learning rate helps as well. This occurs because the problem is a constrained convex optimization. If you go too fast then you can fly off the surface and...